[hieptuanle]

Summary

Code 128 had no way to set the ^BC m parameter. Every export was mode N (Subset B, one symbol per character), and an imported ^BC…,A silently lost its A: the parser kept the field, the preview rendered it at Subset B width, and the next export wrote it back as mode N.

For numeric payloads that costs a lot of label width. 0240018268623 at ^BY2:

	modules	dots
mode N (today)	178	356
mode A	123	246

Changes:

• Barcode1DProps gains an optional mode ('N' | 'A', CODE128_MODES), honoured by Code 128 only. gs1 (m=D) still wins, so a GS1-128 field never emits A.
• Emit: ^BCo,h,f,g,e,A when mode === 'A'; the default stays the bare command, so existing designs export byte-identically.
• Import: ^BC…,A parses into mode: 'A' (new bcAuto field flag, reset alongside bcGs1).
• Canvas: planCode128AutoSymbols (in code128Subset.ts) plans mode A per ISO/IEC 15417 Annex E over Subsets B/C (start in C on a leading run of 4+ digits or a 2-digit payload; mid-payload runs of 4+ switch to C, an odd run leaves its first digit in B; drop back to B for a trailing single digit). buildBwipOptions renders that plan as bwip raw, so the preview width equals the printed width. Payloads with control bytes return null and keep the existing A/B invocation path.
• Panel: an "Auto subset (compact digits)" checkbox on Code 128, hidden while GS1 mode is on. Label added to every locale (including the new ca).
• MCP: code128 props doc lists mode.

Note: Annex E is the reference here, not firmware measurements. When a run has an odd digit count, Zebra's auto mode may leave a different digit outside Subset C. Symbol count and width are the same either way, and so is the decoded data, but bar-for-bar parity with a real ZD-series print is unverified. I don't have a Zebra handy to measure; happy to adjust if you see a difference.

Test plan

• pnpm exec vitest run: 374 files / 6138 tests pass. New packages/core/src/lib/code128Auto.test.ts covers the planner (even/odd leading runs, short runs stay in B, 2-digit payload, mid-payload switches, control-byte fallback), emit with and without A, GS1 winning over A, import of ^BC…,A vs bare ^BC, and the rendered footprint (246 vs 356 dots)
• tsc -b and eslint on the changed files clean
• Used in production behind our vendored copy (subtree of v0.8.1) in a label designer since 2026-10-27

🤖 Generated with Claude Code

[u8array]

Thanks for the PR! Since #593 landed, the canvas's existing ^FD reader should handle mode A as well, while mode derivation and capability handling should move into core.

I'll take it from here.